News
DealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory’s NVIDIA & AMD Enterprise ExpansionDealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory’s NVIDIA & AMD Enterprise Expansion
EveryNeocloud
Mithril

Mithril

A GPU VM marketplace with spot bids and flexible reservations for AI workloads

Mithril, formerly called Foundry, is a US company that sells GPU virtual machines through a marketplace where prices move with supply and demand. It was set up by alumni of Google DeepMind's deep learning team and Stanford's computer science PhD programme, and says teams run more than 10,000 GPU-hours a day on its platform.

Share on WhatsApp

  • Formerly known as Foundry
  • Spot bids priced live by supply and demand
  • Reservations credit back unused hours
  • Teams run 10,000+ GPU-hours a day, the company says
  • CLI built on open-source SkyPilot for portability

What it is

Mithril runs a marketplace of GPU virtual machines aimed at people building AI systems. The company describes its goal as making cloud compute simpler and more widely available, so that AI teams do not have to spend their time working out how much capacity they will need. Pricing is set by supply and demand: customers place bids for spot capacity, and can also take out reservations that guarantee access while crediting back hours they do not use. The company says its users include AI-native startups, research labs, listed companies and academic institutions, and that teams run over 10,000 GPU-hours a day across the platform. It is based in the Palo Alto and San Francisco Bay Area area.

How it came about

The business was previously known as Foundry and now trades as Mithril. It was started by people who came out of the core Deep Learning group at Google DeepMind and the computer science PhD programme at Stanford, with a team spanning machine learning theory, distributed systems and financial modelling. The company's own pages describe its culture in terms of balancing rigidity with flexibility, clear communication and a bias for action.

The compute

Mithril sells GPU virtual machines rather than naming specific accelerator models. Capacity is bought in two ways: spot bids, where a low bid waits for the price to fall and a high bid starts work immediately at whatever the market clears at, and flexible reservations that guarantee access while letting customers pause or return capacity they do not need. The platform offers a hosted Kubernetes control plane, so reserved and spot instances join one cluster and standard kubectl commands work as usual. Its command-line tool builds on the open-source SkyPilot project, so existing task YAML files run unchanged and stay portable across the clouds SkyPilot supports. Cloud buckets from S3, GCS and R2 mount directly into tasks, and every console action is available as a command or API call.

Customers and deals

Mithril says its users cover AI-native startups, research labs, public companies and academic institutions, together running more than 10,000 GPU-hours a day. Arc Institute, the developer of the EVO genomic foundation model, is named as a customer. Its co-founder and CEO says the platform gives researchers the compute they need without procurement friction. The company positions itself as running alongside the clouds customers already use rather than replacing them.

Its place in AI

Mithril presents itself as an 'omnicloud' for AI, a layer that sits next to existing clouds rather than instead of them. Its team includes alumni of Google DeepMind's deep learning group and Stanford's computer science PhD programme. The company page lists advisers and supporters with links to UC Berkeley, Databricks, Covariant AI, Physical Intelligence and a former Google chief executive, though the page does not spell out each person's role. Arc Institute credits the platform with supporting its genomic foundation model work.

Questions people ask about Mithril

What does Mithril sell?

A marketplace of GPU virtual machines for AI work. Capacity is bought either as spot instances priced live by supply and demand, or as flexible reservations that guarantee access and credit back hours you do not use. The company says teams run more than 10,000 GPU-hours a day on the platform.

Who founded the company?

Mithril was founded by alumni from the core Deep Learning team at Google DeepMind and from Stanford's computer science PhD programme. The team covers machine learning theory, distributed systems and financial modelling. The company was previously called Foundry.

How does pricing work?

You set a bid for spot capacity. A low bid means your workload waits until the price drops. A high bid acts as a safety limit and the job starts straight away, paying whatever the market clears at. Flexible reservations guarantee GPU access but let you pause or return capacity you do not need.

Does it replace my existing cloud?

No. Mithril says it runs alongside the clouds you already use. Everything in the console is available as commands and API calls, the hosted Kubernetes control plane accepts your instances into one cluster, and the CLI is built on SkyPilot so existing task files stay portable.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Mithril yet, so the facts on this page come from the sources listed above: mithril.ai/company, mithril.ai/. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryNeocloud wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the provider of the day, new pages on EveryNeocloud, and the conferences and anniversaries of the week ahead.